Described herein is a graphics processing unit (GPU) comprising a first processing cluster to perform parallel processing operations, the parallel processing operations including a ray tracing operation and a matrix multiply operation; and a second processing cluster coupled to the first processing cluster, wherein the first processing cluster includes a floating-point unit to perform floating point operations, the floating-point unit is configured to process an instruction using a bfloat16 (BF16) format with a multiplier to multiply second and third source operands while an accumulator adds a first source operand with output from the multiplier.

Methods, apparatus, systems and articles of manufacture are disclosed facilitate read-modify-write support in a coherent victim cache with parallel data paths. An example apparatus includes a random-access memory configured to be coupled to a central processing unit via a first interface and a second interface, the random-access memory configured to obtain a read request indicating a first address to read via a snoop interface, an address encoder coupled to the random-access memory, the address encoder to, when the random-access memory indicates a hit of the read request, generate a second address corresponding to a victim cache based on the first address, and a multiplexer coupled to the victim cache to transmit a response including data obtained from the second address of the victim cache.

Methods, apparatus, systems and articles of manufacture are disclosed facilitate read-modify-write support in a coherent victim cache with parallel data paths. An example apparatus includes a random-access memory configured to be coupled to a central processing unit via a first interface and a second interface, the random-access memory configured to obtain a read request indicating a first address to read via a snoop interface, an address encoder coupled to the random-access memory, the address encoder to, when the random-access memory indicates a hit of the read request, generate a second address corresponding to a victim cache based on the first address, and a multiplexer coupled to the victim cache to transmit a response including data obtained from the second address of the victim cache.

A caching system including a first sub-cache and a second sub-cache in parallel with the first sub-cache, wherein the second sub-cache includes: line type bits configured to store an indication that a corresponding cache line of the second sub-cache is configured to store write-miss data, and an eviction controller configured to evict a cache line of the second sub-cache storing write-miss data based on an indication that the cache line has been fully written.

A caching system including a first sub-cache and a second sub-cache in parallel with the first sub-cache, wherein the second sub-cache includes: line type bits configured to store an indication that a corresponding cache line of the second sub-cache is configured to store write-miss data, and an eviction controller configured to evict a cache line of the second sub-cache storing write-miss data based on an indication that the cache line has been fully written.

Techniques described herein relate to systems and methods of data storage, and more particularly to providing layering of file system functionality on an object interface. In certain embodiments, file system functionality may be layered on cloud object interfaces to provide cloud-based storage while allowing for functionality expected from a legacy applications. For instance, POSIX interfaces and semantics may be layered on cloud-based storage, while providing access to data in a manner consistent with file-based access with data organization in name hierarchies. Various embodiments also may provide for memory mapping of data so that memory map changes are reflected in persistent storage while ensuring consistency between memory map changes and writes. For example, by transforming a ZFS file system disk-based storage into ZFS cloud-based storage, the ZFS file system gains the elastic nature of cloud storage.

Techniques described herein relate to systems and methods of data storage, and more particularly to providing layering of file system functionality on an object interface. In certain embodiments, file system functionality may be layered on cloud object interfaces to provide cloud-based storage while allowing for functionality expected from a legacy applications. For instance, POSIX interfaces and semantics may be layered on cloud-based storage, while providing access to data in a manner consistent with file-based access with data organization in name hierarchies. Various embodiments also may provide for memory mapping of data so that memory map changes are reflected in persistent storage while ensuring consistency between memory map changes and writes. For example, by transforming a ZFS file system disk-based storage into ZFS cloud-based storage, the ZFS file system gains the elastic nature of cloud storage.

A processing system rinses, from a cache, those cache lines that share the same memory page as a cache line identified for eviction. A cache controller of the processing system identifies a cache line as scheduled for eviction. In response, the cache controller, identifies additional “dirty victim” cache lines (cache lines that have been modified at the cache and not yet written back to memory) that are associated with the same memory page, and writes each of the identified cache lines to the same memory page. By writing each of the dirty victim cache lines associated with the memory page to memory, the processing system reduces memory overhead and improves processing efficiency.

A processing system rinses, from a cache, those cache lines that share the same memory page as a cache line identified for eviction. A cache controller of the processing system identifies a cache line as scheduled for eviction. In response, the cache controller, identifies additional “dirty victim” cache lines (cache lines that have been modified at the cache and not yet written back to memory) that are associated with the same memory page, and writes each of the identified cache lines to the same memory page. By writing each of the dirty victim cache lines associated with the memory page to memory, the processing system reduces memory overhead and improves processing efficiency.

A fleet of query accelerator nodes is established for a data store. Each accelerator node caches data items of the data store locally. In response to determining that an eviction criterion has been met, one accelerator node removes a particular data item from its local cache without notifying any other accelerator node. After the particular data item has been removed, a second accelerator node receives a read query for the particular data item and provides a response using a locally-cached replica of the data item.